# meta-llama/PurpleLlama

Set of tools to assess and improve LLM security.

Repository: https://github.com/meta-llama/PurpleLlama
Canonical: https://ross.abutalabs.com/products/purplellama
Language: Python
License: NOASSERTION
License Family: other
Last push: 2026-08-18T01:15:09+00:00

## Health v2 (maintenance only)
Score: 71/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 98, release rhythm 35, longevity 71
- inputs: {"age_days": 1001, "days_push": 16, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases, no_license
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 4365, forks 772 (observed 2026-08-28T04:08:46.640460+00:00)

## What it is
Purple Llama is Meta's umbrella project of tools and benchmarks for assessing and improving the security of large language models. It includes CyberSecEval cybersecurity evaluations and Llama Guard input/output safeguard models for human-AI conversations.

## Use cases
- evaluate cybersecurity risks of an LLM
- add safety filtering to LLM inputs and outputs
- red team a chatbot for unsafe responses
- benchmark a model for insecure code generation
- moderate human-AI conversations with a classifier

## When to choose
- you need open benchmarks for LLM security risks
- you want a ready-made safeguard model for prompt and response moderation
- you are building responsibly with Llama or other open generative AI models

## When to avoid
- you need general-purpose content moderation outside LLM safety
- you require a permissive license for the safeguard models themselves (they use the Llama Community license)
- you need a hosted SaaS safety API rather than self-run models

## Facets
- artifact type: library
- maturity: active
- function: security, llm-inference, machine-learning, benchmarking, prompt-engineering
- domain: security, large-language-models, artificial-intelligence, developer-tools
- platform: python, cross-platform
- tags: llm-safety, llama-guard, cybersecurity-evals, red-teaming, input-output-safeguards, trust-and-safety, meta

## Member repositories
- meta-llama/PurpleLlama (main) score 71

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:08:46.640460+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-29T18:21:25.268658+00:00, confidence not recorded.
  - readme: https://github.com/meta-llama/PurpleLlama (fetched 2026-08-28T04:08:46.640460+00:00, sha 28af4ecfc234)
- Data as of 2026-08-30T08:39:29.467469+00:00.